Use splitfngr with nlminb

Use nlminb function but pass in a single function that returns both the function and gradient together in a list. Useful when the function and gradient are expensive to calculate and can be calculated faster together than separate.

Usage

nlminb_share(start, fngr, ...)

Arguments

start

Initial values for the parameters to be optimized over. Will be passed to nlminb as start argument.

fngr

A function that returns a list of two elements: the function value and the gradient value.

...

Other arguments passed to nlminb

Value

Result from running nlminb on the given function

Examples

quad_share <- function(x){list(sum(x^4), 4*x^3)}
nlminb_share(start=c(3, -5), fngr=quad_share)
#> $par
#> [1]  1.083723e-12 -1.083081e-12
#> 
#> $objective
#> [1] 2.755427e-48
#> 
#> $convergence
#> [1] 0
#> 
#> $iterations
#> [1] 124
#> 
#> $evaluations
#> function gradient 
#>      132      125 
#> 
#> $message
#> [1] "relative convergence (4)"
#> 

if (FALSE) { # \dontrun{
# Add a sleep amount to show when it can be faster

# Using share
quad_fngr <- function(x){Sys.sleep(.01); list(sum(x^4), 4*x^3)}
system.time(nlminb_share(start=c(3, -5), fngr=quad_fngr))

# Without share
quad_fn <- function(x) {Sys.sleep(.01); sum(x^4)}
quad_gr <- function(x) {Sys.sleep(.01); 4*x^3}
system.time(nlminb(c(3,-5), quad_fn, quad_gr))
} # }